1. Run predeploy migration in db/migrate to remove the unique constraint 2. Deploy new version of code that ignores the column 3. Run post-deploy migration that removes column 4. Remove the part of code that ignores the column
Failure to do step 1 would result in multiple rows being created with null values, which would cause errors for all but one (or zero) insert. The edge case, however, is a little more subtle. Between steps 1 and 2, there's a period where the database constraint doesn't exist, but the old code still expects the constraint. Validating uniqueness in the code can mitigate this, but doesn't ensure consistency.